CN117494871A - Ship track prediction method considering ship interaction influence - Google Patents

Ship track prediction method considering ship interaction influence Download PDF

Info

Publication number
CN117494871A
CN117494871A CN202311275303.XA CN202311275303A CN117494871A CN 117494871 A CN117494871 A CN 117494871A CN 202311275303 A CN202311275303 A CN 202311275303A CN 117494871 A CN117494871 A CN 117494871A
Authority
CN
China
Prior art keywords
ship
data
track
time
model
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202311275303.XA
Other languages
Chinese (zh)
Inventor
张新宇
黄瑞宁
刘震生
姜玲玲
李高才
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Dalian Maritime University
Original Assignee
Dalian Maritime University
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Dalian Maritime University filed Critical Dalian Maritime University
Priority to CN202311275303.XA priority Critical patent/CN117494871A/en
Publication of CN117494871A publication Critical patent/CN117494871A/en
Pending legal-status Critical Current

Links

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q10/00—Administration; Management
    • G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/042—Knowledge-based neural networks; Logical representations of neural networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/044—Recurrent networks, e.g. Hopfield networks
    • G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/045—Combinations of networks
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/04—Architecture, e.g. interconnection topology
    • G06N3/0464—Convolutional networks [CNN, ConvNet]
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
    • G06N3/00—Computing arrangements based on biological models
    • G06N3/02—Neural networks
    • G06N3/08—Learning methods
    • G06N3/084—Backpropagation, e.g. using gradient descent

Landscapes

  • Engineering & Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Theoretical Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Computational Linguistics (AREA)
  • Evolutionary Computation (AREA)
  • Software Systems (AREA)
  • Mathematical Physics (AREA)
  • General Engineering & Computer Science (AREA)
  • Computing Systems (AREA)
  • Molecular Biology (AREA)
  • General Health & Medical Sciences (AREA)
  • Data Mining & Analysis (AREA)
  • Biophysics (AREA)
  • Biomedical Technology (AREA)
  • Artificial Intelligence (AREA)
  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Business, Economics & Management (AREA)
  • Human Resources & Organizations (AREA)
  • Economics (AREA)
  • Strategic Management (AREA)
  • Tourism & Hospitality (AREA)
  • Game Theory and Decision Science (AREA)
  • Development Economics (AREA)
  • Quality & Reliability (AREA)
  • Operations Research (AREA)
  • Marketing (AREA)
  • Entrepreneurship & Innovation (AREA)
  • General Business, Economics & Management (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)

Abstract

The invention relates to a ship track prediction method considering ship interaction influence, which comprises the following steps: preprocessing ship AIS historical data to obtain a ship time sequence track data set; constructing a ship space network relation graph according to the interactive influence relation among ships, and acquiring a ship space relative relation; calculating a ship approach effect according to the ship speed and the ship relative distance to acquire a ship time relation; constructing a neural network ship track prediction model based on GCN and GRU; constructing a self-defined mixed loss function of the loss function; setting network training parameters and outputting the track of the ship in a future period. The model of the invention combines the graphic neural network and the gating circulation unit to be applied to the field of ship track prediction, effectively improves the scientificity of the prediction result by utilizing the sensitivity of the graphic structure to the space and the sensitivity of the gating circulation unit to the time, and ensures the accuracy of the track prediction result. The method can provide more effective technical support for the scenes such as ship behavior analysis, water situation awareness and the like.

Description

Ship track prediction method considering ship interaction influence
Technical Field
The invention relates to the technical field of ship track prediction, in particular to a ship track prediction method considering ship interaction influence.
Background
The ship track prediction difficulty of considering other ship interaction relations in a water area with frequent traffic conflict, large flow and complex traffic condition is high. The traditional ship track prediction method generally utilizes a ship kinematic model to predict the microscopic motion behavior of the ship, but the traditional kinematic model is easy to be influenced by an original data error, and modeling difficulty is high under a complex environment considering dynamic factors such as wind, waves, flow and the like based on physical modeling. The neural network can process uncorrelated characteristics, can perform feasible and good-effect analysis on a large-scale data source in a relatively short time, has the characteristics of simple calculation, easy understanding, strong interpretation and the like, and is more applied to ship track prediction research. The scholars model the ship track by using a Long Short-Term Memory network (LSTM), and excavate the space-time distribution rule of the track in a data driving mode to predict the future track of the ship. But neglects the interaction of the ship with other surrounding ships in complex traffic environments such as narrow water areas and ports. The result of the trajectory prediction is too idealized and lacks interpretability to solve the actual trajectory prediction problem.
Disclosure of Invention
In view of the defects of the prior art, the invention provides a ship track prediction method considering the ship interaction influence, the graph neural network and the gating circulation unit are combined to be applied to the field of ship track prediction, the scientificity of a predicted result is effectively improved by using the sensitivity of a graph structure to space and the sensitivity of the gating circulation unit to time, and the accuracy of the track predicted result is ensured.
The invention adopts the following technical means:
a ship track prediction method considering ship interaction influence comprises the following steps:
s1: performing data preprocessing on the classified original ship AIS track data, performing data integrity analysis, space-time track synchronization and generating corresponding ship time sequence data and a ship feature matrix;
s2: establishing an interaction module from two angles of time approaching efficiency and space distance to quantitatively consider the interaction between ships, and representing the preprocessed ship time series data by using a ship space relation graph to regenerate an adjacent matrix of the ship space relation;
s3: constructing a ship track prediction model of a combined neural network based on a graph convolution network and a gating circulation unit;
s4: taking the ship characteristic matrix and the adjacency matrix as model input, taking ship time sequence data as output of the model to train the ship track prediction model of the combined neural network, and carrying out back propagation by using a self-defined loss function, and setting a network parameter training model, wherein the loss function is as follows:
wherein y is k Represents the kth tag value, y k * Represents the kth predictive value, χ is the L2 regularization term coefficient, specified in training as 0.001, ω t The parameters of the model are m parameters in total, and j is the total number of labels.Is the absolute error of longitude and latitude variation->Error of track included angle, ">For the L2 regular term, the model complexity is reduced by reducing the numerical size of the weight.
S5: and predicting the ship time sequence data of the next time period by using the trained combined neural network ship track prediction model, and outputting predicted track points of the ship in the next time period according to the predicted ship time sequence data.
Further, the step S1 of data preprocessing includes the following sub-steps:
s1.1: labeling the space position information in the ship AIS data through longitude and latitude, and converting the longitude and latitude coordinates into the ink card bracket plane coordinates;
s1.2: cleaning ship track data, including course, abnormal speed cleaning, abnormal longitude and latitude cleaning and abnormal MMSI data, wherein:
aiming at ship data in a navigation section, the course speed cleaning calculates the course speed change rate of adjacent track points, calculates the upper and lower quarters of the change rate in five minutes before and after the calculation to judge whether the course speed change of the point is abnormal, judges whether the course of the point is abnormal by giving a course change rate and a quartile range, and corrects the course by using the average value of the upper and lower ten digits of all records in five minutes before and after the record if the course change rate and the quartile range are judged to be abnormal;
the abnormal cleaning of longitude and latitude comprises deleting the data with missing longitude and latitude and beyond the normal longitude and latitude range in the navigation track data, and deleting track points with the same longitude and latitude coordinates and different time;
s1.3: interpolation is carried out in an interpolation mode of the inherent attribute of AIS data, the ship kinematic characteristics are combined before linear interpolation is carried out, two groups of predicted values are obtained, and then the time difference from the missing value to the observation point is used as a weight to carry out linear interpolation;
s1.4: then for each track, using every 6 consecutive track points as one sample data, taking 5 as time step and the last as future track point, n-5 sample data can be extracted from a track sequence with n consecutive track points, thereby obtaining a ship AIS data set, and dividing the ship AIS data set into a training set and a testing set according to the proportion of 8:2.
Further, in the step S2, an interaction module is established from two angles of time approaching efficiency and space distance to quantitatively consider the interaction between ships, and the preprocessed ship time series data is represented by a ship space relation diagram, which includes:
s2.1: the ship relation network is represented by G, and the non-weight graph G (V, E) = { V 1 ，V 2 ，……，V N -representing a topology of a ship network, wherein N represents the number of nodes, E represents a set of edges, adjacency matrix a represents connections between ships, a E R N×N ；
S2.2, quantitatively considering the mutual influence between ships by utilizing two angles of time approaching efficiency and space distance to obtain a ship characteristic matrix X epsilon R N×P P represents the number characteristics of the ship node attributes;
s2.3: and normalizing the ship characteristic matrix data by adopting a dispersion normalization method, and converting all characteristic data into data between [0,1 ].
Further, in the step S3, constructing a ship track prediction model based on a graph convolution network and a gate control circulation unit, including:
s3.1: the method comprises the steps of designing a GCN network to extract a spatial dependency relationship of a ship relationship network topological structure, firstly inputting a determined adjacency matrix A and a feature matrix X into a GCN model, wherein the GCN model is a filter constructed in a Fourier domain, the filter acts on nodes of a graph, two layers of GCN models are constructed through spatial features among first-order neighborhood capture nodes of the filter, and the GCN model capture spatial dependency relationship is constructed through superposition of a plurality of convolution layers;
s3.2: designing a GRU network structure to extract time relation of a ship time sequence, wherein the GRU model structure comprises a reset gate and an updateThe door is used for storing the ship information in a hidden state and controlling the last ship state information h t-1 And the reset gate is responsible for neglecting the influence on the ship in the ship space diagram at the previous moment.
Further, the step S4 includes:
the adjacent matrix A and the feature matrix X are used as inputs, the GCN is utilized to extract the space dependence of the track sequence according to the ship relationship network topology structure, and a Droupout layer is connected behind each time convolution module and used for preventing the problem of overfitting;
and inputting the obtained ship track time sequence of the spatial features into a gating unit model, wherein the gating unit model takes the hidden state at the moment t-1 and the current ship state as inputs to obtain the ship motion state at the moment t, captures the time dynamic change features, uses a loss function to perform back propagation training model, and finally obtains a result through a full-connection layer.
Further, the ship track prediction in the step S5 may be expressed as:
on the premise of giving a ship network relation topological graph G and a feature matrix X, a mapping function f is learned, and then ship positions at the future time T are calculated and predicted, wherein the following formula is shown:
[X t+1 ,...,X t+T ]＝f(G；(X t-n ,...，X t-1 ，X t ))
where n represents the historical time series length, T represents the predicted time series length, h t-1 Is the hidden state at the moment t-1, X t For the traffic information at time t, R t For resetting the gate, for controlling the degree of neglect of the state information at the previous moment, U t To update the gate for controlling the extent to which the state information at the previous time is brought into the current state, C t Is the memory content stored at the time t, h t The output state at time t.
Compared with the prior art, the invention has the following advantages:
the ship track prediction method considering the ship interaction influence provided by the invention provides a new thought of considering the interaction relation between ships and the ship time and space relation in complex scenes, the method considers the motion data and the interaction with the surrounding environment, the position prediction error is effectively reduced, the requirement of ship track prediction in the development environment in the large age can be met, in addition, the traditional graph neural network is further improved, the sensitivity of the graph neural network to the space topology and the gating circulation unit to time is applied, so that the prediction result has better interpretation, the prediction result has better performance in the aspect of training time sequence related problems, and the accuracy of track prediction is better.
Drawings
In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings that are required in the embodiments or the description of the prior art will be briefly described, and it is obvious that the drawings in the following description are some embodiments of the present invention, and other drawings may be obtained according to the drawings without inventive effort to a person skilled in the art.
FIG. 1 is a flow chart of a ship track prediction method considering the ship interaction effect.
FIG. 2 is a schematic view of the overall processing structure of the present invention.
Fig. 3 is a schematic diagram of a GCN network structure according to an embodiment of the present invention.
Fig. 4 is a schematic diagram of a GRU network structure according to an embodiment of the present invention.
Fig. 5 is a schematic diagram of an adjacency matrix according to an embodiment of the present invention.
Detailed Description
In order that those skilled in the art will better understand the present invention, a technical solution in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings in which it is apparent that the described embodiments are only some embodiments of the present invention, not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the present invention without making any inventive effort, shall fall within the scope of the present invention.
Referring to fig. 1 to 2, a ship track prediction method considering the interactive influence of a ship includes the steps of:
s1: performing data preprocessing on classified original ship AIS track data, performing data integrity analysis, space-time track synchronization and generating corresponding ship time sequence data P t ＝{MMSI，Log t ，Lat t Wherein MMSI (Water Mobile service identification code), log (longitude), lat (latitude) and ship characteristic matrix X εR X.P P represents the characteristic of the ship node property (such as the ship length, the ship width and the like), N represents the node number, lm/Bm represents the aspect ratio of the first ship, and the larger the size difference of the ships among different ships is, the larger the interaction effect among the ships is. Preferably, in this embodiment, the ship feature matrix is:
specifically, the method comprises the steps of firstly obtaining AIS original data, preprocessing the AIS original data and generating corresponding ship time series data. The method specifically comprises the following steps:
s1.1: the ship AIS data is marked with space position information through longitude and latitude, and the coordinates of the longitude and the latitude are converted into the plane coordinates of the mercator for subsequent calculation:
wherein, longitude and latitude of the track point are marked as (alpha, beta), and rectangular coordinates in the converted ink card bracket coordinate system are (x, y), r 0 The method is defined as a reference dimension circle radius, q is defined as an equivalent dimension, a in the formula is an ellipse long radius of the earth, and e is a first eccentricity of the ellipsoid.
S1.2, main problems of ship track data include single record single attribute, multiple associated attributes of single record and repetition, deletion and error of multiple associated attributes of multiple records. The ship AIS data cleaning mainly comprises course, speed abnormal cleaning, longitude and latitude abnormal cleaning and MMSI abnormal data cleaning.
The anomaly data includes:
A. and (3) repeating the AIS data, namely, in order to avoid the situation of data dead zones, a plurality of AIS receiving systems are usually arranged, and a part of overlapping areas are formed by adjacent systems, so that the phenomenon of repetition occurs when the AIS data of the ship is realized. Therefore, an operation of data deduplication is required. If the numerical values of all fields in the dynamic and static data of the same MMSI are completely consistent, deleting the repeated part and only reserving one unique data record.
B. Heading, speed anomaly data: deleting data outside a heading range (0-360 degrees), deleting AIS data with the navigational speed of negative value or less than 1 maritime/hour, and repairing by adopting a mean value interpolation mode if the navigational speed is changed.
C. Longitude and latitude abnormal data: deleting the data with missing longitude and latitude and beyond the normal longitude and latitude range in the navigation track data, and deleting the track points with the same longitude and latitude coordinates and different time.
D. MMSI anomaly data: and (3) eliminating data (generally 9 digits) of which the marine mobile service identification code (MMSI) does not accord with the specified length, wherein the MMSI is inconsistent with the ship type and the ship length.
Interpolation is carried out in an interpolation mode of the inherent attribute of AIS data, the ship kinematic characteristics are combined to obtain two groups of predicted values before linear interpolation is carried out, and then the time difference from the missing value to the observation point is used as a weight to carry out linear interpolation.
Wherein v and w are the speed and heading of the observation point, respectively, t (time stamp of the point to be inserted. PairWeighted average is performed on the two predicted sets of values, and the weights are t i (and t) 1 、t 2 The smaller the time interval, the greater the weight that the set of predictors takes in the final interpolated value, determined by the time difference between them. The specific definition is as follows:
w 1 and w 2 For the weight of A, B, the interpolation coordinates after weighted average are:
and S2, establishing an interaction module from two angles of time approaching efficiency and space distance to quantitatively consider the interaction between ships, and representing the preprocessed ship time series data by using a ship space relation graph to regenerate an adjacent matrix of the ship space relation.
Specifically, the interaction module is established by utilizing the two angles of the time approaching efficiency and the space distance to quantitatively consider the interaction between ships, and mainly comprises the following steps:
s2.1: judging the mutual influence relation between ships by using the time approaching efficiency:
wherein:the relative distance and relative speed of the two vessels, respectively. If the time approaching efficiency is more than or equal to 0, the two ships are shown to be in a scattered trend, and under the condition, the interaction influence between the ships is not considered, the two ships do not form a connecting edge, and the adjacent matrix is expressed by 0; if the approach rate is less than 0, showing that the ships are in a converging trend, and forming a connecting edge between the two ships.
S2.2: the stable motion state and the motion state change stage of the ship are separated when the track is segmented, but considering that the traffic condition of a complex water area is complex, not all ships can take proper avoidance actions, the mutual influence of the ships is a continuous and dynamic change process, and the characteristics of behavior of the ships can be known, under the condition that the avoidance actions are taken, the ship spacing is firstly reduced and then increased, under the condition that the avoidance actions are not taken, the ship spacing can be firstly reduced and then increased, and two ships can collide, so that the ship spacing is always reduced. Therefore, it is considered that when the distance between the two vessels approaches to a certain extent, there is a potential interaction between the two vessels regardless of the motion states of the two vessels, and therefore, the vessel interaction threshold D is set.
S2.3: the ship AIS data after pretreatment and cleaning is represented by a ship space relation diagram, a filter is constructed in a Fourier domain, the filter captures space characteristics among ships to capture potential time characteristics among different time series data, and the ship AIS data is converted into an array form in codes, so that an adjacent matrix A of a time similarity matrix is generated. As shown in fig. 5, the adjacency matrix a of the present invention is preferably an adjacency matrix a which mainly indicates whether there is a mutual influence between ships, and is marked as 1 for the mutual influence and 0 for the non-mutual influence.
Then obtaining a ship characteristic matrix X epsilon R according to the attribute data of the ship N×P P represents the number characteristic of the ship node attribute.
S3: and (3) entering an adjacent matrix A of the ship characteristic matrix X and the time similarity matrix into a graph convolution neural network, and acquiring ship relation net puff structure information by utilizing the graph neural network to acquire space characteristics. Time series data with spatial features are obtained.
Wherein X is a feature matrix, A is an adjacency matrix,representing the computational process, ++>Representing adjacency matrix and self-join procedure, +.>Readiness matrix,/->W o And W is 1 Representing the weight matrix of the first layer and the second layer, σrelu () represents the activation function.
S4: inputting the time sequence matrix with the space characteristics into a gating circulation unit model, taking the hidden state of t-1 and the current traffic information as inputs to obtain the traffic state at the moment t, and outputting the ship time sequence data with the time space characteristics.
Update door: z is Z t ＝σ(X t W xz +H t-1 W hz +b z )
Reset gate: r is R t ＝σ(X t W xr +H t-1 W hr +b r )
W and b represent weights and deviations during training.
When the model is trained, the ship characteristic matrix and the adjacent matrix are used as model inputs, ship time sequence data is used as the output of the model to train the combined neural network ship track prediction model, a self-defined loss function is used for carrying out back propagation, and a network parameter training model is set, wherein the loss function is as follows:
wherein y is k Represents the kth tag value, y k * Represents the kth predictive value, χ is the L2 regularization term coefficient, specified in training as 0.001, ω t The parameters of the model are m parameters in total, and j is the total number of labels.Is the absolute error of longitude and latitude variation->Error of track included angle, ">For the L2 regular term, the model complexity is reduced by reducing the numerical size of the weight.
S5: predicting ship AIS time series data of the next time period by using the trained model, and outputting a predicted track point P of the ship of the next time period at the t+1th second according to the predicted ship AIS time series data t1 The ship trajectory prediction may be expressed as: on the premise of giving a ship network relation topological graph G and a feature matrix X, a mapping function f is learned, and then ship positions at the future time T are calculated and predicted, wherein the following formula is shown:
[X t+1 ,...,X t+T ]＝f(G；(X t-n ,...，X t-1 ，X t ))
referring to fig. 3, the GCN model established in step S3 includes an input layer, a hidden layer, softMax, and an output layer. The GCN model constructs a filter in the Fourier domain. The filter acts on nodes of a ship relation network diagram, space features among ships are captured through a first-order neighborhood, and then a GCN network model is constructed through a plurality of convolution layers, wherein the formula is as follows:
where H is the spatial feature extracted for each layer, A is the adjacency matrix,i is an identity matrix, representing an adjacency matrix and a self-connection process, sigma is a nonlinear activation function,/-a>Is->Is a degree matrix of (2).
The resulting time series with spatial features is expressed as: p (P) space ＝(p t1 ,p t2 ,...,p tm ) i
Wherein p is tm Is a time stamp t m Corresponding position point, i represents MMSI number, P space Representing a time series with spatial specifications.
Referring to fig. 4, in step S3, the gated loop cell model is one of a loop neural network, adapted to solve the long term memory and back-propagation gradient problem. Which includes a reset gate and an update gate. And acquiring the time characteristics of the time series of the acquired space characteristics by inputting the input information at the moment of the current year and the hidden state at the last moment.
The resulting time series with spatial and temporal features is expressed as: p (P) space-time ＝(p t1 ,p t2 ,...,p tm ) i
Wherein p is tm Is a time stamp t m Corresponding position point, i represents MMSI number, P space-time Representing a time series with spatial and temporal characteristics.
In addition, the Dropout method is adopted to deal with the over fitting problem which may occur to the neural network model, and the threshold value is set to be 0.5.
Inputting the ship sequence data with the obtained space and time characteristics into a GCN () (graph convolutional networks) -GRU (Gated Recurrent Unit) network model trained in the step 7, setting an Epoch, the number of temporary nodes and the position Size, repeating the experiment for a plurality of times, predicting the ship AIS time sequence data of the next time period, and outputting the predicted track point P of the ship in the next time period at the t+1th second according to the predicted ship AIS time sequence data t1 And obtaining a prediction result and a prediction error.
Finally, it should be noted that: the above embodiments are only for illustrating the technical solution of the present invention, and not for limiting the same; although the invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical scheme described in the foregoing embodiments can be modified or some or all of the technical features thereof can be replaced by equivalents; such modifications and substitutions do not depart from the spirit of the invention.

Claims (6)

1. The ship track prediction method considering the ship interaction influence is characterized by comprising the following steps of:
s1: performing data preprocessing on the classified original ship AIS track data, performing data integrity analysis, space-time track synchronization and generating corresponding ship time sequence data and a ship feature matrix;
s2: establishing an interaction module from two angles of time approaching efficiency and space distance to quantitatively consider the interaction between ships, and representing the preprocessed ship time series data by using a ship space relation graph to regenerate an adjacent matrix of the ship space relation;
s3: constructing a ship track prediction model of a combined neural network based on a graph convolution network and a gating circulation unit;
s4: taking the ship characteristic matrix and the adjacency matrix as model input, taking ship time sequence data as output of the model to train the ship track prediction model of the combined neural network, and carrying out back propagation by using a self-defined loss function, and setting a network parameter training model, wherein the loss function is as follows:
wherein y is k Represents the kth tag value, y k * Represents the kth predictive value, χ is the L2 regularization term coefficient, specified in training as 0.001, ω t The parameters of the model are m parameters in total, and j is the total number of labels. S5: using trainedAnd predicting ship time sequence data of the next time period by the combined neural network ship track prediction model, and outputting predicted track points of the ship in the next time period according to the predicted ship time sequence data.
2. A ship track prediction method taking into account the interaction of ships according to claim 1, wherein the step S1 of data preprocessing comprises the following sub-steps:
s1.1: labeling the space position information in the ship AIS data through longitude and latitude, and converting the longitude and latitude coordinates into the ink card bracket plane coordinates;
s1.2: cleaning ship track data, including course, abnormal speed cleaning, abnormal longitude and latitude cleaning and abnormal MMSI data, wherein:
aiming at ship data in a navigation section, the course speed cleaning calculates the course speed change rate of adjacent track points, calculates the upper and lower quarters of the change rate in five minutes before and after the calculation to judge whether the course speed change of the point is abnormal, judges whether the course of the point is abnormal by giving a course change rate and a quartile range, and corrects the course by using the average value of the upper and lower ten digits of all records in five minutes before and after the record if the course change rate and the quartile range are judged to be abnormal;
the abnormal cleaning of longitude and latitude comprises deleting the data with missing longitude and latitude and beyond the normal longitude and latitude range in the navigation track data, and deleting track points with the same longitude and latitude coordinates and different time;
s1.3: interpolation is carried out in an interpolation mode of the inherent attribute of AIS data, the ship kinematic characteristics are combined before linear interpolation is carried out, two groups of predicted values are obtained, and then the time difference from the missing value to the observation point is used as a weight to carry out linear interpolation;
s1.4: then for each track, using every 6 consecutive track points as one sample data, taking 5 as time step and the last as future track point, n-5 sample data can be extracted from a track sequence with n consecutive track points, thereby obtaining a ship AIS data set, and dividing the ship AIS data set into a training set and a testing set according to the proportion of 8:2.
3. The ship track prediction method according to claim 1, wherein in the step S2, the interaction module is established from two angles of time approaching efficiency and space distance to quantitatively consider the interaction between ships, and the preprocessed ship time series data is represented by a ship space relation diagram, comprising:
s2.1: the ship relation network is represented by G, and the non-weight graph G (V, E) = { V 1 ，V 2 ，……，V N -representing a topology of a ship network, wherein N represents the number of nodes, E represents a set of edges, adjacency matrix a represents connections between ships, a E R N×N ；
S2.2, quantitatively considering the mutual influence between ships by utilizing two angles of time approaching efficiency and space distance to obtain a ship characteristic matrix X epsilon R N×P P represents the number characteristics of the ship node attributes;
s2.3: and normalizing the ship characteristic matrix data by adopting a dispersion normalization method, and converting all characteristic data into data between [0,1 ].
4. The ship track prediction method considering the ship interaction effect according to claim 1, wherein in the step S3, constructing a combined neural network ship track prediction model based on a graph convolution network and a gate control circulation unit comprises:
s3.1: the method comprises the steps of designing a GCN network to extract a spatial dependency relationship of a ship relationship network topological structure, firstly inputting a determined adjacency matrix A and a feature matrix X into a GCN model, wherein the GCN model is a filter constructed in a Fourier domain, the filter acts on nodes of a graph, two layers of GCN models are constructed through spatial features among first-order neighborhood capture nodes of the filter, and the GCN model capture spatial dependency relationship is constructed through superposition of a plurality of convolution layers;
s3.2: the GRU network structure is designed to extract the time sequence time relation of the ship, the GRU model structure comprises a reset door and an update door, the update door stores ship information in a hidden state, and the control is carried outLast ship status information h t-1 And the reset gate is responsible for neglecting the influence on the ship in the ship space diagram at the previous moment.
5. The ship track prediction method considering the ship interaction effect according to claim 4, wherein the step S4 comprises:
the adjacent matrix A and the feature matrix X are used as inputs, the GCN is utilized to extract the space dependence of the track sequence according to the ship relationship network topology structure, and a Droupout layer is connected behind each time convolution module and used for preventing the problem of overfitting;
and inputting the obtained ship track time sequence of the spatial features into a gating unit model, wherein the gating unit model takes the hidden state at the moment t-1 and the current ship state as inputs to obtain the ship motion state at the moment t, captures the time dynamic change features, uses a loss function to perform back propagation training model, and finally obtains a result through a full-connection layer.
6. The ship track prediction method considering the ship interaction effect according to claim 5, wherein the ship track prediction in step S5 can be expressed as:
on the premise of giving a ship network relation topological graph G and a feature matrix X, a mapping function f is learned, and then ship positions at the future time T are calculated and predicted, wherein the following formula is shown:
[X t+1 ,...,X t+T ]＝f(G；(X t-n ,...，X t-1 ，X t ))
where n represents the historical time series length, T represents the predicted time series length, h t-1 Is the hidden state at the moment t-1, X t For the traffic information at time t, R t For resetting the gate, for controlling the degree of neglect of the state information at the previous moment, U t To update the gate for controlling the extent to which the state information at the previous time is brought into the current state, C t Is the memory content stored at the time t, h t Output at time tStatus of the device.
CN202311275303.XA 2023-09-28 2023-09-28 Ship track prediction method considering ship interaction influence Pending CN117494871A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202311275303.XA CN117494871A (en) 2023-09-28 2023-09-28 Ship track prediction method considering ship interaction influence

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202311275303.XA CN117494871A (en) 2023-09-28 2023-09-28 Ship track prediction method considering ship interaction influence

Publications (1)

Publication Number Publication Date
CN117494871A true CN117494871A (en) 2024-02-02

Family

ID=89671577

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202311275303.XA Pending CN117494871A (en) 2023-09-28 2023-09-28 Ship track prediction method considering ship interaction influence

Country Status (1)

Country Link
CN (1) CN117494871A (en)

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117893575A (en) * 2024-03-15 2024-04-16 青岛哈尔滨工程大学创新发展中心 Ship motion prediction method and system with self-attention mechanism integrated by graph neural network
CN117893575B (en) * 2024-03-15 2024-05-31 青岛哈尔滨工程大学创新发展中心 Ship motion prediction method and system with self-attention mechanism integrated by graph neural network

Cited By (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN117893575A (en) * 2024-03-15 2024-04-16 青岛哈尔滨工程大学创新发展中心 Ship motion prediction method and system with self-attention mechanism integrated by graph neural network
CN117893575B (en) * 2024-03-15 2024-05-31 青岛哈尔滨工程大学创新发展中心 Ship motion prediction method and system with self-attention mechanism integrated by graph neural network

Similar Documents

Publication Publication Date Title
Liang et al. Spatial-temporal aware inductive graph neural network for C-ITS data recovery
US10909446B2 (en) Systems and methods for selecting global climate simulation models for training neural network climate forecasting models
Nguyen et al. A multi-task deep learning architecture for maritime surveillance using AIS data streams
CN109142171B (en) Urban PM10 concentration prediction method based on feature expansion and fusing with neural network
Cirstea et al. EnhanceNet: Plugin neural networks for enhancing correlated time series forecasting
CN114220271A (en) Traffic flow prediction method, equipment and storage medium based on dynamic space-time graph convolution cycle network
CN109255505A (en) A kind of short-term load forecasting method of multi-model fused neural network
Venskus et al. Unsupervised marine vessel trajectory prediction using LSTM network and wild bootstrapping techniques
CN115862324A (en) Space-time synchronization graph convolution neural network for intelligent traffic and traffic prediction method
CN116502161A (en) Anomaly detection method based on dynamic hypergraph neural network
CN117494871A (en) Ship track prediction method considering ship interaction influence
Huang et al. Ship trajectory anomaly detection based on multi-feature fusion
Rani et al. Gridlock Prediction using a Gated Recurrent Neural Network for Smart Transport Systems
Rezaei Nezhad et al. Optimized Deep Networks Structure to Improve the Accuracy of estimator algorithm in Deep Networks learning
Khampuengson Machine learning methods for detecting and correcting data errors in water level telemetry systems
Nezhad et al. Optimized deep networks structure to improve the accuracy of estimator algorithm in deep networks learning.
CN116610919A (en) Space time sequence prediction method based on cyclic graph algorithm sub-neural network
CN116629115A (en) Bidirectional data driving ship track prediction method and system based on attention mechanism
CN118072873A (en) Air quality prediction method
CN115169502A (en) Deep learning technology-based fishing boat behavior classification method and system
CN116248524A (en) Communication network situation prediction model and method for ocean vessels
Agrawal PREPRINT Yet another Case-Based Reasoning system application
Agrawal PREPRINT A Case-Based Reasoning system application
Agrawal A Case-Based Reasoning system application
Ress Modeling the ocean with Multi-Agent Systems

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination